Image Segmentation Research Based On Active Contour Models

Image segmentation has always been a fundamental problem and complex task in the field of image processing and computer vision. Image segmentation aims at partition the original image into several homogeneous characteristic regions and isolating the interesting region. The Active Contour Model (ACM) methods based on Partial Differential Equation (PDE) are proposed to address a wide range of image segmentation problems. Due to combining low-level image information with high-level prior knowledge, it has shown unique advantages and comprehensive applicabilities contrast with the traditional image segmentation methods.This dissertation aims to research on image segmentation with active contour models based on Partial Differential Equation. In this thesis, we first make a review of the present image segmentation algorithms. Then, the classification of active contour models and the relative mathematical knowledge are introduced. Some classical active contour models are detailed. Finally, we mainly focus on the research of active contour models based on Partial Differential Equation.In this thesis, we focus our research on the following aspects:1. To solve the problem of the initial contour sensitivity, we define a new data-fitting energy incorporating local Gaussian intensity means and variances. Then, the energy minimization is achieved by an interleaved level set evolution. In the resulting curve evolution that minimizes the energy functional, the motion of the contour is driven by a local intensity fitting force that incorporates the local intensity information and the Gaussian parameters. This force plays a key role in overcoming the initial contour sensitivity and extracting the object boundaries. In addition, the regularity of level set function is preserved to ensure accurate computation and avoids the expensive re-initialization of evolving level set function. Experimental results show the advantages of our method in terms of robustness and efficiency over some well-known active contour models. 2. We define a new data-fitting energy which incorporates both global and local image intensity information for image segmentation. The energy functional of the proposed model consists of three parts:global term, local term and regularization term. Due to using the local image information, the images with intensity inhomogeneity can be efficiently segmented. Because of the global term, it allows more flexible initialization of contours. Then the regularity of the level set function is preserved to ensure accurate computation and avoids expensive reinitialization of the evolving level set function. Moreover, we use the Gaussian filtering to regularize our level set function, which keeps the level set function smooth. The proposed method can be applied to the field of image processing and computer vision with promising results.3. We propose a novel global active contour model for image segmentation. By incorporating the bias function and the local image data, the images with intensity inhomogeneity can be efficiently segmented. By considering the global information, it allows more flexible initialization of contours and solves the problem of falling into the local minimum. Then the regularity of the level set function is preserved to ensure accurate computation and avoids expensive reinitialization of the evolving level set function. Experiments demonstrate that our proposed method makes a significant improvement in segmenting synthetic, natural and medical images.